Appearance
Nucleotide Transformer v2 50M
Nucleotide Transformer v2 50M is a small DNA/RNA language model used for sequence embeddings and lightweight genomic representation tasks.
What it does
The model turns nucleotide sequences into numerical embeddings. Embeddings are vectors that summarize sequence patterns in a way downstream tools can compare, cluster, or score.
When to use it
Use this model when you want a faster, lighter sequence embedding option. It is appropriate for short windows, demos, and early pipeline design before moving to larger models.
Inputs in Liatir
- FASTA/FA/FNA file, or an inline DNA/RNA sequence.
- For variant effect workflows: reference FASTA plus
.vcfor.vcf.gzvariants. - Molecule type: DNA or RNA.
- Maximum token/window length.
Outputs
Liatir can produce:
- per-sequence embeddings;
- variant effect scores for small local VCF/VCF.GZ batches;
- BED tracks for genome viewer inspection;
- JSON/CSV summaries;
- basic metrics such as sequence count and embedding size;
- provenance with model ID, revision, runtime, input, and parameters.
Hardware and installation
This model can run on CPU for small batches. GPU or Apple Metal through PyTorch can be faster when available.
Liatir installs the model through a managed Python runtime using PyTorch and Transformers.
Limits and cautions
The model license is non-commercial. Check whether your use case is allowed before using it in commercial or restricted work.
Embeddings are not direct biological conclusions. They are numerical representations that need downstream interpretation.
For variant effect scoring, a higher embedding delta means the model representation changed more. It does not mean the variant is automatically pathogenic or clinically important.